Time optimality for systems with multidimensional control and vector moment min-problem (Q2185761)

The subject is the time optimal problem for systems \[ x'(t) = Ax(t) + Bu(t) \, , \qquad x(t) \in\mathbb{R}^n \, , \ u(t) \in \Omega \subset\mathbb{R}^r \tag{1} \] with \(r \ge 2.\) The constraint set is the hypercube \(|u_j| \le 1.\) Under rank conditions on the matrices \(B, AB, \dots, A^{n-1}B\) and a general position condition that, if not satisfied, may lead to nonuniqueness, Pontryagin's maximum principle says that the controls take only the values \(\pm 1;\) if the matrix \(A\) has only real eigenvalues the number of switching point of each control does not exceed \(n - 1.\) That reduces the problem to finding the switching points, which can be done explicitly in some low dimensional examples but is in general very complicated, especially in high dimensions. The authors reduce the time optimal problem for (1) to a \textit{vector moment problem} as follows. If \(\Phi(t)\) is the fundamental matrix of the homogeneous system and the control \(u(t)\) steers \(x_0\) to \(0\) in time \(\theta\) then \[ \Phi (\theta)x_0 + \Phi(\theta) \int_0^{\theta} \Phi^{-1}(t) B(t) u(t) dt = 0 \] so the problem becomes that of finding the minimum time \(\theta\) and the vector function \(u(t)\) that satisfies the constraints and \[ \int_0^\theta \Psi (t)u(t) dt = x_0 \] where \(\Psi(t) = - \Phi(t)^{-1} B(t).\) The two cases considered are \textit{zero spectrum} (only one eigenvalue \(\lambda = 0)\) and \textit{real spectrum} under assumptions including linearly independent eigenvectors. The final results are nonlinear systems for explicit computation of the switching points. The numerics are illustrated in an example.